Spaces:
Running
Running
Fix loading the text model
Browse files
app.py
CHANGED
@@ -1,21 +1,22 @@
|
|
1 |
import os
|
|
|
2 |
|
3 |
import matplotlib.pyplot as plt
|
4 |
import numpy as np
|
5 |
import pandas as pd
|
6 |
import streamlit as st
|
7 |
-
from transformers import CLIPProcessor
|
8 |
|
9 |
from medclip.modeling_hybrid_clip import FlaxHybridCLIP
|
10 |
|
11 |
|
12 |
-
@st.
|
13 |
def load_model():
|
14 |
-
model
|
15 |
-
|
16 |
-
return model,
|
17 |
|
18 |
-
@st.
|
19 |
def load_image_embeddings():
|
20 |
embeddings_df = pd.read_hdf('feature_store/image_embeddings_large.hdf', key='emb')
|
21 |
image_embeds = np.stack(embeddings_df['image_embedding'])
|
@@ -64,7 +65,7 @@ elif ex4_button:
|
|
64 |
|
65 |
|
66 |
image_list, image_embeddings = load_image_embeddings()
|
67 |
-
model,
|
68 |
|
69 |
query = st.text_input("Enter your query here:", value=text_value)
|
70 |
dot_prod = None
|
@@ -78,8 +79,8 @@ if st.button("Search") or k_slider:
|
|
78 |
else:
|
79 |
with st.spinner(f"Searching ROCO test set for {query}..."):
|
80 |
k = k_slider
|
81 |
-
inputs =
|
82 |
-
|
83 |
query_embedding = model.get_text_features(**inputs)
|
84 |
query_embedding = np.asarray(query_embedding)
|
85 |
query_embedding = query_embedding / np.linalg.norm(query_embedding, axis=-1, keepdims=True)
|
@@ -91,4 +92,4 @@ if st.button("Search") or k_slider:
|
|
91 |
for img_path, score in zip(matching_images, top_scores):
|
92 |
img = plt.imread(os.path.join(img_dir, img_path))
|
93 |
st.image(img, width=300)
|
94 |
-
st.write(f"{img_path} ({score:.2f})"
|
|
|
1 |
import os
|
2 |
+
import token
|
3 |
|
4 |
import matplotlib.pyplot as plt
|
5 |
import numpy as np
|
6 |
import pandas as pd
|
7 |
import streamlit as st
|
8 |
+
from transformers import CLIPProcessor, AutoTokenizer
|
9 |
|
10 |
from medclip.modeling_hybrid_clip import FlaxHybridCLIP
|
11 |
|
12 |
|
13 |
+
@st.cache_resource
|
14 |
def load_model():
|
15 |
+
model = FlaxHybridCLIP.from_pretrained("flax-community/medclip-roco", _do_init=True)
|
16 |
+
tokenizer = AutoTokenizer.from_pretrained('allenai/scibert_scivocab_uncased')
|
17 |
+
return model, tokenizer
|
18 |
|
19 |
+
@st.cache_resource
|
20 |
def load_image_embeddings():
|
21 |
embeddings_df = pd.read_hdf('feature_store/image_embeddings_large.hdf', key='emb')
|
22 |
image_embeds = np.stack(embeddings_df['image_embedding'])
|
|
|
65 |
|
66 |
|
67 |
image_list, image_embeddings = load_image_embeddings()
|
68 |
+
model, tokenizer = load_model()
|
69 |
|
70 |
query = st.text_input("Enter your query here:", value=text_value)
|
71 |
dot_prod = None
|
|
|
79 |
else:
|
80 |
with st.spinner(f"Searching ROCO test set for {query}..."):
|
81 |
k = k_slider
|
82 |
+
inputs = tokenizer(text=[query], return_tensors="jax", padding=True)
|
83 |
+
# st.write(f"Query inputs: {inputs}")
|
84 |
query_embedding = model.get_text_features(**inputs)
|
85 |
query_embedding = np.asarray(query_embedding)
|
86 |
query_embedding = query_embedding / np.linalg.norm(query_embedding, axis=-1, keepdims=True)
|
|
|
92 |
for img_path, score in zip(matching_images, top_scores):
|
93 |
img = plt.imread(os.path.join(img_dir, img_path))
|
94 |
st.image(img, width=300)
|
95 |
+
st.write(f"{img_path} ({score:.2f})")
|